Notes from the team building FuturOne. We write when we have something to say — about agent architecture, zero-retention infrastructure, and how we know whether an agent's work is any good.
An agent that ships deliverables needs a sharper bar than "looks plausible" — and for a code review agent, "passes tests" measures the code, not the review. How we score artifacts against versioned rubrics, calibrate confidence against acceptance outcomes, and what the 91% acceptance number actually counts.
Read the post →Most vendors offer a retention policy — a promise to delete data from a system that stores it by default. We built a system with nothing to delete, and it collapsed enterprise security reviews from months to about two weeks. Here is how it works, and what it honestly costs us.
Read the post →Every product page says "agent" now; most of them ship a chat loop with a longer system prompt. The real difference is a planner–orchestrator–verifier loop over tools and durable state — and you cannot prompt your way into one.
Read the post →Prefer watching to reading? Watch an agent run live, or follow smaller releases on the changelog.